摘要
为了能够在复杂的工业环境中抓取平面型工件,提出一种图割法与形状先验模型结合的工件图像分割方法,并且对工件的位姿信息进行测量.首先,建立先验形状,提出基于最小包围矩形法将工件的形状模板与目标工件人工分割形状进行配准,得到先验形状.为了保证分割结果的准确性,采用单一的先验形状.图割模型中加入了目标形状先验知识.其次,通过自适应调整形状先验项的权重系数,使得图割法的能量函数形状先验项自适应于被分割的图像.第三,本文可以采用形状先验方法分割一幅图像中的多个工件,并且能够计算吸盘的最优抓取位置.最后,采用结构光视觉系统采集工件的点云信息,拟合工件平面,确定工件法向量,得到工件的抓取姿态.实验结果表明,本文算法能够适应遮挡、光照变化的工件图像,同时也能够分割复杂环境中的目标工件;平面型工件抓取位姿的计算结果有效,可以应用于遮挡、反光、复杂干扰背景条件下的工件抓取作业.
A new shape prior segmentation method based on graph cuts is used to segment workpiece images and measure the workpiece posture for grasping workpieces in cluttered industry scene. Firstly, a prior shape is built. Minimum bounding rectangle method is proposed to register the workpiece shape model and the manual shape of the target workpiece to get the prior shape. In order to ensure the segmentation accuracy, a single prior shape is used. The target shape prior knowledge is added to the graph cut model. Secondly, the weight of the shape prior term is adjusted in a self-adaptive manner, so that the shape prior term of the energy function in graph cut method becomes adaptive to the image to be segmented. Thirdly, multiple workpieces in a image can be segmented by the shape prior method. Meanwhile, the optimal position of suction cup for grasping the workpiece is determined. Finally, the structured light vision system is used to acquire the point cloud of the workpiece. The plane of the workpiece is fitted and the normal vector is determined. Thus, the grasping orientation is obtained. The effectiveness of the proposed approach is demonstrated on the workpiece segmentation in the scene with occlusion, light variation and cluttered background. The posture of the planar workpiece acquired through calculation is accurate, and can be applied to the grasping operation in conditions of occlusion, reflection and complicated background. © 2017, Science Press. All right reserved.
出处
《机器人》
EI
CSCD
北大核心
2017年第1期99-110,共12页
Robot
基金
国家863计划(2013AA041002-1)
国家科技支撑计划(2015BAK06B01)
国家自然科学基金(61403372
61403374)
